Data Primitives: What Is Not Done

The three data primitives are at different stages. This page exists so that is knowable from one place rather than discovered mid-integration. Last reviewed: 2026-08-03

​

One-line status

​

One vendor, one token

Everything except Bluesky and Hacker News runs on Apify actors behind a single APIFY_API_TOKEN. There is no per-provider subscription to procure and no optional code path that behaves differently depending on which keys exist. A suspended or lapsed account fails loudly rather than degrading. Email verification and the company tech-stack lookup both used to swallow every provider exception, which meant a suspended account returned “unverified” and “no technologies” indefinitely and looked healthy. Account-level faults now propagate; only genuinely transient actor failures degrade a record, and the record says which datasets contributed.

​

Social Data

Bluesky and Hacker News are verified against the live APIs: real posts, working URLs, and a real comment thread round-tripped through collect. Not done:

• X and Reddit are unverified live. The actors are wired, priced and covered by hermetic tests, but no run has completed while the Apify account is suspended. When the People adapter first met a real API it had three genuine bugs — a query shape matching nothing, a mishandled empty result, and a field with the wrong type. Assume the same class of issue.
• Reddit is batch-only, by measurement. 53–219 seconds for five posts on live runs, so both sync entry points refuse it. Not a limitation to be lifted; a shape.
• include_comments is unsupported on Bluesky and X, and refused rather than billed.
• prefer_official is refused for X and Reddit — there is no official-API adapter, and claiming one would be worse than saying so.

​

People

Not done:

• Unverified against the live provider while the account is suspended.
• Email-only and domain-only enrich do not resolve. The enrich actor keys on LinkedIn identity, so linkedin, or name plus company, are what work. Both other shapes stay on the request contract and are refused with a hint rather than answered empty.
• No phone numbers, by design — the provider’s person-level phone fields are personal numbers, which the B2B boundary withholds.
• match_likelihood is never populated. The current provider returns a profile or nothing and scores neither.
• No provider waterfall. One provider, one attempt. A miss returns null, so match rates are that provider’s match rates.
• No provider evaluation has been run. Coverage and match rate against a real target list — the numbers that decide whether this primitive is any good — are unmeasured.

​

Company Data

Not done:

• Unverified against the live provider while the account is suspended.
• The funding overlay is best-effort. It is keyed on a Crunchbase slug derived from the company name, and derivation is inexact — a company whose slug diverges will not match. A miss degrades the record to firmographics, and sources reports what contributed.
• Search cannot filter on funding, stage or headcount. Those filters are refused before any actor starts rather than silently dropped, because the search dataset has no such fields.
• recent_news, employee_growth_6m and tags are never populated. No actor in the set returns news, headcount growth needs two observations over time, and market category tags are deliberately not conflated with a real detected tech_stack.

​

Across all three

• Async batch scope is unconfirmed. A 1,000-record batch returns raw records in one payload, which may exceed what a provider licence permits as “embedded in product functionality” versus reselling data. The cap is PRIMITIVE_BATCH_MAX pending written confirmation.
• Pricing is now measured, not derived. Every Apify-backed meter bills 3× the run’s own provider_usd, so there is no per-record list price left to drift and no per-meter margin to review: a slow, expensive run bills more and a cheap one bills less. The projection from the actor table survives only as the pre-flight reservation and as the fallback when a run reports no usage at all.
• PRIMITIVES_MOCK serves fixtures through the real adapters and still charges credits. It fails closed outside a local dev/test process, so it cannot be set in staging or production.
• No operator UI. API and CLI only; the dashboard shows the primitives in its catalog but has no surface for these calls.

​

Commercial and legal, not code

None of these are fixed by shipping software:

• Provider agreement permitting redistribution through a customer-facing API
• Subprocessor disclosure naming Apify in Naive’s tenant DPA
• Privacy-policy line disclosing the use of third-party B2B data providers
• Data-broker registration — California, Texas, Oregon and Vermont maintain registries, and selling access to personal data about people with no direct relationship is close to the definition. A company-level obligation with statutory penalties, independent of any provider’s own compliance.